Commodity Price Forecasting in the International Market: Using a Proposed Ensemble Approach, Time Series and Machine Learning Models

Abstract

This research paper presents a comprehensive analysis of regression and time-series models for predicting commodity prices in the international market, focusing on Brent Oil, US Soybeans, and US Wheat. The study evaluates the accuracy and effectiveness of these models using performance metrics such as Root Mean Squared Error (RMSE) and R2 score. Additionally, a novel hybrid model is proposed, incorporating Random Forest, Decision Tree, Gradient Boosting, and further refined using a meta model - Linear Regression. The results of the analysis indicate that the hybrid model outperforms the majority of the traditional models and time-series models in terms of forecasting accuracy. Time-series models, specifically ARIMA and Prophet, demonstrate impressive performance in in-sample prediction. However, challenges were encountered with the LSTM model, which exhibited it to be computationally intensive and required careful parameter selection. To address these limitations, the research proposes potential avenues for improvement, although it acknowledges that the accuracy of predictions cannot be guaranteed. The findings of this study provide valuable insights for policymakers, investors, and risk analysts, offering a deeper understanding of the performance of different models in predicting commodity prices.